使用 keras ImageDataGenerator - AttributeError: 'DirectoryIterator' object has no attribute 'argmax' 时如何获取分类报告（F1 分数）

Question

Since I do not have the training set and label separated when using Keras ImageDataGenerator but rather I rely on the folder structure. How can I get the classification report, ie how can I get/calculate the Precision, Recall, and F1?

This is what I have tried and I get the error that is complaining that the train_generator has not argmax. How can I solve this?

AttributeError: 'DirectoryIterator' object has no attribute 'argmax'

train_datagen = ImageDataGenerator(
    rescale=1. / 255,
)

train_generator = train_datagen.flow_from_directory(
    train_dir, 
    target_size=(150, 150),  
)

predIdxs = model.predict(train_generator)

predIdxs = np.argmax(predIdxs, axis=1)

print(classification_report(train_generator.argmax(axis=1), predIdxs,
                            target_names=["class 1", "class 2"]))

Answer 1

This solution worked for me since if you use flow_from_directory then you must use predict_generator method on model and speaking about sklearn.metrics.classification_report , ytrue and ypred should be 1-d numpy array hence used train_generator.labels

train_datagen = ImageDataGenerator(
    rescale=1. / 255,
)

train_generator = train_datagen.flow_from_directory(
    train_dir, 
    target_size=(150, 150), shuffle = False,   # Must
)

predIdxs = model.predict_generator(train_generator)

predIdxs = np.argmax(predIdxs, axis=1)

print(classification_report(train_generator.labels, predIdxs,
                            target_names=["class 1", "class 2"]))